Recognition means “to know again” or “recall to mind”. Visual Recognition is then concerned with assigning semantic category labels to visual inputs, such as images or videos. Our research at the Center has tackled several visual recognition tasks, such as studying the decomposition of objects into parts for understanding the object, exploring contour detectors that exploit high-level semantic information, and studying the human pose through the pose of the arms. More recently, research on Visual Recognition at the Center has focused on (1) localizing objects in challenging contexts, such as detecting four-leafed clovers, and (2) fixing the brittleness of state-of-the-art Visual Recognition systems against malicious image perturbations.
